The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation in references/seedance.md instructs the agent to perform a git clone of an untrusted repository (github.com/wwwzhouhui/jimeng-free-api-all.git) and subsequently execute a Python script (scripts/logout-sessions.py) from that repository. This provides a direct path for executing remote code from an unverified third-party source.
[EXTERNAL_DOWNLOADS]: The skill requires cloning a third-party repository and installing external dependencies (playwright) which are not associated with a trusted organization or the skill author.
[COMMAND_EXECUTION]: In SKILL.md, troubleshooting instructions include piping the output of a curl command directly to the python3 interpreter (curl -s http://localhost:8000/v1/models | python3 -m json.tool). While used for JSON formatting, this pattern is flagged as a high-risk vector for remote code execution.
[DYNAMIC_EXECUTION]: The skill uses the source command to load environment variables from .env files located in project or user directories. If an attacker gains write access to these files, they can inject arbitrary shell commands that will be executed in the agent's context when the skill is loaded.
[CREDENTIALS_UNSAFE]: The skill manages highly sensitive Jimeng session IDs by storing them in local .env files and instructing the agent to source these files. This exposes credentials directly to the shell environment and increases the risk of credential theft.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted user prompts and processes external media URLs without appropriate boundary markers or sanitization.
Ingestion points: prompt parameter in all generation endpoints, images URL array in compositions, and file uploads in references/seedance.md.
Boundary markers: None present to distinguish embedded instructions from user data.
Capability inventory: Access to Bash (executing docker and curl) and Write (writing files to the system).
Sanitization: No evidence of sanitization or filtering for user-supplied prompt strings or external media content.

nby-jimeng-api